Joyce told the audience that the attacks demonstrated something the security community has been debating for years: AI agents can find and exploit vulnerabilities with a relentlessness that human attackers cannot match. "It freakin' worked," he said, describing how the AI-automated approach allowed attackers to probe systems continuously without fatigue.

The original Anthropic report, published in late 2025, detailed how Chinese intelligence operatives had used Claude to automate reconnaissance, identify attack surfaces, and generate exploit code. The revelation split the infosec community, with some seeing it as proof that AI safety measures are insufficient and others arguing it simply accelerated techniques that were already well known.

Joyce framed the divide as revealing about the industry itself. Those focused on defence saw a terrifying escalation. Those focused on offence saw an inevitable evolution. Both were looking at the same data and drawing different conclusions — hence the Rorschach test analogy.

The talk comes as the relationship between AI companies and government agencies remains fraught, with the Pentagon recently cutting ties with Anthropic despite being close to a deal.